Regen

Regen

Incident reader + correlation extractor for Fluidify Regen (self-hosted, AGPLv3 on-call tool). Regen ships a REST API but no MCP server, so this agent reads incidents first-class and extracts the keys to pivot into SigNoz telemetry. Symmetric with the signoz skill: regen answers "what broke and when"; signoz answers "why," from the traces/metrics/logs.

What You Get

• Read incidents and ingested alerts (with their labels) and timelines
• regen_correlation_keys -- extracts service.name, correlation labels (chain, role, address, intent_id, ...), a time window, and a ready-to-run SigNoz filter hint
• Optional write tools (ack / resolve / update) behind an explicit flag

The mini-axol deployment

• REGEN_BASE_URL = http://mini-axol.tail9b2ce8.ts.net:3302 -- Tailscale must be up (a connection failure is almost always a down tailnet).
• Runs in open mode behind the Tailscale ACL (Regen OSS falls through to open mode when no cookie/SAML is configured), so no secret is needed -- only REGEN_BASE_URL. Guardrail: do not create a Regen user or enable SAML, or open mode closes and the wrapper breaks.
• Alerts originate from SigNoz on the same host and carry service.name, chain, role, etc. -- the labels that make the SigNoz pivot exact.

MCP tools

Read (always available):

Tool	Description
regen_list_incidents	List incidents; filter by status, severity, created_after/before, limit
regen_get_incident	One incident + linked alerts (with labels) + timeline
regen_list_alerts	Ingested alerts; filter by status, source, limit
regen_correlation_keys	Extract service.name + labels + time window + SigNoz filter hint

Write (only when REGEN_ENABLE_WRITE=1 or regen serve --write):

Tool	Description
regen_ack_incident	Acknowledge an incident
regen_resolve_incident	Resolve an incident (optional summary)
regen_update_incident	Update status / severity / summary

Write tools mutate live on-call state -- confirm intent before calling them.

Correlation workflow (incident -> OTel)

1. A SigNoz alert (e.g. SolverGasCriticalOptimism) fires on mini-axol, POSTs to Regen's webhook, and opens an incident whose labels carry service.name, chain, role, etc.

2. regen_correlation_keys(<id>) pulls those labels + the incident's time window and emits a SigNoz filter hint, e.g.:

   service.name IN (riddler-balance-exporter, riddler-production)
     AND chain='optimism' AND role='solver'
     AND time>=<start> AND time<=<end>
   

3. Feed that hint into the signoz MCP -- use the window as explicit start/end (unix ms) so telemetry lines up with when the alert fired: signoz_search_logs, signoz_aggregate_traces, signoz_query_metrics (the riddler_* metrics). This closes the loop from incident back to OTel.

The /signoz:incident_triage alertId=<id> prompt automates the SigNoz side once you have the alert rule id.

CLI Usage

regen incidents --severity critical   # list (status/severity filters)
regen incident 42                     # inspect one: alerts + timeline
regen alerts --source prometheus      # list ingested alerts
regen correlate 42                    # the correlation-keys step
regen serve                           # MCP server (stdio); add --write for mutations

Configure

Variable	Required	Purpose
REGEN_BASE_URL	yes	Regen instance URL (origin only; client appends /api/v1)
REGEN_SESSION_COOKIE	no	oi_session cookie value (local-auth deploys only)
REGEN_API_TOKEN	no	Authorization: Bearer (forward-compatible)
REGEN_ENABLE_WRITE	no	1/true to expose the write tools

Point Claude Code at a wrapper that injects REGEN_BASE_URL from 1Password, mirroring the signoz wrapper:

{ "mcpServers": { "regen": { "command": "regen-mcp-wrapper.sh", "args": [] } } }

Install

cd agents/regen
pip install -e ".[dev]"

See also

• signoz -- the telemetry side of the loop; where correlation keys get queried
• observability-designer -- design SLOs and alert rules that open these incidents
• sentinel -- on-chain contract monitoring (a different alert source)